Journal of Experimental Psychology: General
2003, Vol. 132, No. 3, 400 – 418
Copyright 2003 by the American Psychological Association, Inc.
0096-3445/03/$12.00 DOI: 10.1037/0096-3445.132.3.400
Modulation of Word-Reading Processes in Task Switching
Michael E. J. Masson and Daniel N. Bub
Todd S. Woodward
University of Victoria
University of British Columbia
Jason C. K. Chan
University of Victoria
The authors examined modulation of the simple act of word naming induced by the conflict arising when
that task competes with color naming in a task-switching paradigm. Subjects alternated between naming
a word printed in black and naming the color of a stimulus in 2 conditions. In the incongruent condition,
the colored stimulus was an irrelevant word generating conflict, and in the neutral condition, color was
carried by a row of asterisks. Subjects took substantially longer to name a word printed in black in the
incongruent condition, implying a form of suppression. This modulation of the word-naming response
was adaptive in that it led to more efficient color naming. The modulation effect was replicated using
phoneme detection instead of word naming but not with lexical decision or visual comparison, implicating a phonological encoding process.
resolving the conflict that is generated when a stimulus has attributes relevant to the two currently active task sets between
which the subject is switching. For example, resolving the conflict
inherent in the act of naming the color in which a word is printed
may lead to the suppression of word-reading processes, which may
then have consequences on a subsequent trial when word reading
is demanded. In this article, we examine the idea that this form of
suppression may constitute an essential component of the cost
associated with switching between tasks.
Task-switching costs arise when subjects are required to shift
between at least two different tasks on a sequence of trials and are
particularly apparent when each target stimulus is bivalent;
namely, it invites the application of either of the two tasks. In
comparison to performance on a series of trials involving a single
task, performance in trials that are part of a sequence of two
alternating tasks is significantly worse (e.g., Allport et al., 1994;
Jersild, 1927; Spector & Biederman, 1976). The cost of switching
between tasks can be reduced but not eliminated by allowing
subjects substantial time (on the order of seconds) to prepare for an
upcoming task switch (e.g., Meiran, 1996; Rogers & Monsell,
1995). The reduction in switch cost achieved by providing subjects
with preparation time is generally accepted as evidence for executive control processes. But the residual switch cost––the cost
remaining after adequate preparatory time is granted––is a component of switch cost that is the target of some controversy.
In one view, even the residual switch cost is ascribed to an
aspect of executive control, whereby a new task cannot be fully
engaged until it is externally cued by a task-relevant stimulus
(Rogers & Monsell, 1995). A rival account of residual switch cost
is that disengagement of a completed task invokes inhibition of
that task set to enable the switch to a new task (Mayr & Keele,
2000). This backward inhibition makes it more difficult to return
to the inhibited task set when required by a subsequent task switch.
Yet another account is that switch cost is the result of the combined effects of (a) inadvertent and inappropriate cuing of the
Responding to stimuli that invite the application of different,
competing processes requires specialized control mechanisms to
resolve conflict. For example, naming the color of a written word
entails avoiding the habitual tendency to read the word, and
subjects are substantially slower to name colors under this condition compared with a condition in which they are required to name
the color of a neutral stimulus, such as a row of Xs (e.g., Klein,
1964). This type of conflict is particularly prevalent in situations
where subjects rapidly switch from one task to another and the
stimulus on each trial affords both the currently demanded task and
the previously executed one (e.g., Allport, Styles, & Hsieh, 1994;
Meiran, 2000; Meiran, Chorev, & Sapir, 2000). Switching between
tasks under these circumstances undoubtedly involves a form of
executive control operating through a central executive or a supervisory attentional system (Baddeley, Chincotta, & Adlam,
2001; Logan & Gordon, 2001; Norman & Shallice, 1986). In
addition, however, there is a need to consider the consequences of
Michael E. J. Masson, Daniel N. Bub, and Jason C. K. Chan, Department of Psychology, University of Victoria, Victoria, British Columbia,
Canada; Todd S. Woodward, Department of Psychology, University of
British Columbia, Vancouver, British Columbia, Canada.
Jason C. K. Chan is now at the Department of Psychology, Washington
University.
This research was supported by research grants from the Natural Sciences
and Engineering Research Council of Canada to Michael E. J. Masson and to
Daniel N. Bub. Order of authorship for the first two authors is arbitrary.
Thanks are extended to Seth Bloom, Shaina Hood, Erica McCollum, Camelia
Pakzad-Vaezi, Neil Pegram, and Lucas Riedl for assistance with data collection. We also thank Iring Koch, Ulrich Mayr, and Nachshon Meiran for helpful
comments on an earlier version of this article.
Correspondence concerning this article should be addressed to either
Michael E. J. Masson or Daniel N. Bub, Department of Psychology,
University of Victoria, P.O. Box 3050 STN CSC, Victoria, British Columbia V8W 3P5, Canada. E-mail: mmasson@uvic.ca or dbub@uvic.ca
400
MODULATION OF WORD READING
previous task by the current stimulus and (b) suppression of an
irrelevant task on trial N that now becomes relevant on trial N ⫹ 1
(Allport et al., 1994; Allport & Wylie, 2000).
The controversy regarding how best to interpret residual switch
costs is not yet resolved. Our goal is not to adjudicate between
different accounts of residual switch costs but to explore the
general idea that suppression arising from conflict between two
operations invited by a bivalent stimulus is an important component of switch costs. Conflict of this nature can arise when a
bivalent stimulus consists of two arbitrarily coupled dimensions,
each affording a different response. For example, in the Rogers
and Monsell (1995) experiments, a bivalent stimulus consisted of
a digit–letter compound, and subjects were cued to respond to one
or the other element on a given trial. In this case, conflict arises
when the irrelevant task set is active by virtue of having been
performed recently and is cued by an aspect of the stimulus
display. Conflict can also occur when one element of a bivalent
stimulus inherently evokes a habitual response that is incompatible
with the desired response to the other element of the stimulus. The
Stroop color-naming task is a classic example of this situation.
Here, the strong tendency to identify a written word invariably
interferes with the task set of naming the color in which an
incongruent color word is presented.
For both of these potential sources of conflict, we suggest that
processing operations associated with the irrelevant task evoked by
a bivalent stimulus are suppressed in the course of responding to
the cued task (see also Allport & Wylie, 2000). If the previously
irrelevant stimulus aspect is then cued to produce a task switch on
a subsequent trial, suppression on trial N of the operations required
for the task on trial N ⫹ 1 will lead to slower responding.
Suppression occurring as part of task-conflict resolution when
responding to a bivalent stimulus is distinct from the concept of
backward inhibition developed by Mayr and Keele (2000). Recall
that backward inhibition is invoked by the requirement to engage
in a task switch and is applied to a task after it is performed. In
contrast, the suppression mechanism that we consider is applied to
processes associated with the irrelevant task and arises because of
the need to resolve the conflict that occurs when two tasks compete
and only one is to be performed. This competition and ensuing
suppression may be especially likely to occur if one task is more
habitual than the other and the subject is required to perform the
less habitual or nondominant task (e.g., naming the color of a
printed word). In a study consistent with this idea, Allport et al.
(1994) reported asymmetrical switch costs, finding larger costs
when subjects switched from color naming to word naming than
vice versa.
To examine the contribution of suppression independent of
other components of switch costs, we constructed a situation in
which a stimulus (e.g., a word printed in color) affords two
possible, competing tasks, only one of which is to be executed.
Conflict is particularly strong if the word refers to a color, but
substantial conflict also occurs even when words do not refer to
colors (Burt, 2002; Klein, 1964; Monsell, Taylor, & Murphy,
2001). We assume that to resolve such task competition, some
aspect or aspects of the nonselected task must be suppressed,
particularly if that task is the more habitual or dominant one.
Therefore, first, one stimulus in a sequence of trials could be a
word printed in color, where the task is to name the color. In this
situation, we assume that subjects would tend to suppress some
401
process involved in the competing, habitual task of word naming.
Second, the next task performed after a task switch would be the
one that was avoided (and that we assume was suppressed) before
the switch. Moreover, because we are interested in directly measuring suppressive effects, we wish to avoid a situation in which
the previous task set (color naming) would be evoked by the
current stimulus. Therefore, the stimulus following the bivalent
item must not contain an element relevant to the previous task.
Thus, after being presented with a colored word for color naming,
subjects must name a word presented in black, when black is not
among the colors used on color-naming trials. We assume that this
constraint on the word-naming stimulus reduces the possibility that
a color-naming response will be cued by the word. Accordingly,
any impact on word reading is unlikely to be the outcome of
conflict induced by the stimulus cuing the previous task set of
color naming.
We arranged color-naming and word-naming tasks as follows.
Each trial consisted of two events (see Figure 1). In the first event,
a word printed in black was presented for the task of naming aloud.
This event was immediately followed by a second event consisting
of a colored stimulus presented for the task of naming the color. In
conditions in which suppression was to be evoked, the color
stimulus was a word. In our experiments, words were never the
names of colors, nor were they strongly associated with color
concepts. Nevertheless, to produce a color-naming response, subjects would have to overcome competition from the potential
response to the word carrying the color (Burt, 2002; Klein, 1964;
Figure 1. Illustration of events for two trials in neutral and incongruent
trial blocks. Stimuli appearing in gray actually were presented in one of
five colors.
402
MASSON, BUB, WOODWARD, AND CHAN
Monsell et al., 2001). We assume that the resolution of this
competition is the source of the suppression that we wish to
investigate. A relatively long pause after the color-naming response and before the start of the next trial was used, allowing the
subject to prepare for the next trial. Any preparatory processes
involved as part of executive control over task performance should
have had ample time to be carried out before the arrival of the
word-naming stimulus on the next trial. By having the wordreading event occur first in each trial, we are in a position to
examine the suppressive effect of color naming on subsequent
word reading, after a substantial interval.
In addition, we eliminated aspects of residual switch costs that
are unrelated to suppression in the following way. In one block of
trials, referred to as the incongruent block, color-naming stimuli
consisted of a word printed in color. On these trials, it is assumed
that word-naming processes are suppressed. Thus, each trial began
with a word-naming event (a word printed in black) and ended
with a color-naming event (a word printed in color). In another
block of trials, referred to as the neutral block, color-naming
stimuli consisted of a row of uniformly colored asterisks. On these
trials, the first event was again word naming (a word printed in
black) and the second event was again color naming (a row of
colored asterisks). An example of each type of trial is shown in
Figure 1.
The critical comparison we make is between word-naming
response times in the incongruent block and word-naming response times in the neutral block. This comparison is taken as a
relatively pure assessment of the influence of suppression (occurring on trial N) on subsequent word reading (occurring on trial N ⫹
1). To reiterate, both the neutral and the incongruent blocks include a substantial delay between color-naming and word-naming
events, allowing ample time for preparation, and we are comparing
word-naming response times under two conditions, which both
require a switch from color naming to word naming. However,
word reading in the incongruent block occurred after the resolution
of a conflict occasioned by naming the color in which an unrelated
word was printed. In the neutral block, word naming occurred in
the absence of conflict.
Our approach contrasts with other methods of examining the
effects of switching in alternating tasks. These methods involve
either (a) comparing a pure block of a single task with a block of
trials involving the same task alternating with another task (the
pure–alternating block method) or (b) alternating runs in which at
least two different tasks are carried out, with each task repeated a
number of times before a switch (the alternating-runs method).
When comparing pure blocks of trials, evidence for a switch cost
consists of less efficient task performance in the alternating block.
There is a disadvantage to the pure–alternating block method in
that only in the alternating block do subjects have to maintain two
task sets in working memory (Fagot, 1994). An advantage to using
the alternating-runs method is that both task sets must be maintained when subjects perform switch and nonswitch trials. The
measure of switch cost in this case is provided by a comparison
between performance on the first trial after a switch with performance on subsequent nonswitch trials.
For both the pure–alternating block method and the alternatingruns method, however, there remains the potential influence of at
least one factor that is not directly associated with switching
between tasks. In particular, we suggest that whenever a bivalent
stimulus is used and conflict resolution is required to make a
selective response, task-switch effects include enduring consequences of this resolution. If conflict resolution leads to the suppression of processing operations that must then be invoked after
a task switch (Allport & Wylie, 2000), then switch costs will
include the cost of overcoming this suppression.
Our method allows us to directly examine the impact of suppression on task performance that can be assessed independent of
switch costs. The critical comparison is between performance in
blocks of trials where switching occurs between two univalent
stimuli (neutral block) and performance in blocks of trials where
switching occurs between a univalent stimulus and a bivalent
stimulus (incongruent block). Both blocks, then, require a task
switch on every trial (word naming followed by color naming).
The crucial difference between the neutral and incongruent blocks
is the use of a bivalent stimulus (a colored word) in the colornaming task. We assume that in the incongruent block, color
naming of a bivalent stimulus requires conflict resolution leading
to the suppression of word naming. If this suppression carries over
to the next word-naming event, then we expect to see slower word
naming than in the neutral block (where color naming involves a
univalent stimulus).
Suppression may have an item-specific component as well as a
more general influence operating at the level of a particular task.
For example, Allport and Wylie (2000) examined word-reading
performance on color words that had served as stimuli in a prior
Stroop color-naming task (i.e., color words printed in a conflicting
color). They also examined word-reading performance on additional color words that subjects had never encountered as Stroop
stimuli. Interference relative to a baseline condition was observed
for both sets of words, although a larger effect was found for words
that had appeared in the Stroop task. These results suggest that
suppression, if it exists, is not confined only to the words that
actually were present as competitors during color naming. Of
course, given that only color words were used throughout, it is
possible that these general effects are due to priming between
conceptually related items (a form of semantic interference; see,
e.g., Yee, 1991).
One of our goals with the research reported here was to determine whether task suppression that may result from conflict in
Stroop-like situations applies to specific items or generalizes
across item types. Suppression of word-reading processes during a
color-naming task may generalize beyond the words that actually
appear in the color-naming task and beyond items conceptually
related to those words. It is well known that interference in color
naming can occur even when color is carried by words that are not
themselves related to color concepts (Dalrymple-Alford, 1972;
Fox, Schor, & Steinman, 1971; Klein, 1964; Monsell et al., 2001).
Therefore, it should be possible to observe effects on word naming
even when the words do not refer to color and are unrelated to
those colors presented on the color-naming task.
In addition, we consider another fundamental question raised by
the notion of task suppression: Does suppression apply to entire
task sets (Gilbert & Shallice, 2002) or to specific component
processes (Meiran et al., 2000)? In the case of switching from
naming a color to naming a word, researchers typically have been
satisfied to construe competition as occurring between whole task
sets (e.g., the task set for color naming and the task set for word
naming). The tendency to theorize at the level of whole tasks rather
MODULATION OF WORD READING
than the level of component processes has been criticized before.
Jacoby (1991) argued against the use of direct and indirect tests of
memory as a means of examining conscious and unconscious
influences because tasks cannot be assumed to be process pure. He
developed an alternative approach based not on dissociations between tasks but on dissociations between component processes.
Neuropsychology and modern aphasiology have benefited considerably in the last few decades from the insistent admonishment by
cognitive scientists that the effects of brain damage on functional
systems cannot be adequately described in terms of global tasks
like reading or color naming. Given such prior awareness of the
limits inherent in the attempt to understand cognitive organization
using global tasks, it is ironic that many current attempts to specify
the nature of switch costs can be faulted for the same emphasis on
tasks rather than cognitive processes as the conflicting elements.
The question of item specificity is addressed in Experiment 1 by
the use of either (a) a small set of repeatedly presented words that
appeared both in color-naming and in word-naming events or (b)
a large set of words, none of which were ever repeated. Because
we use unrelated noncolor words in the large, unrepeated set, we
cannot ascribe any effects obtained on word naming to itemspecific suppression or priming among related items. Rather, such
effects would be due to suppression of either the general task of
word naming or some specific processing mechanism. To address
the issue of process specificity, we also examined suppression
effects on tasks other than word naming. One task involved visual
comparison of nonlinguistic characters, and this allowed us to
determine whether suppression generated by the color-naming task
applies generally to the encoding of visual forms and/or to the
production of a response to a stimulus. An alternative possibility is
that suppression affects a specific stage in the process of encoding
an orthographic stimulus into a phonological or an articulatory
code. If the locus of suppression is at a relatively early stage, such
as orthographic encoding, then any word-reading task should show
suppression.
Alternatively, the suppression may affect some later stage of
encoding involving the construction of a phonological code, in
which case suppression effects would not be seen in word-reading
tasks that can be executed without relying on a phonological code.
The lexical-decision task is a candidate for this purpose given that
Hino and Lupker (2000) and Waters and Seidenberg (1985) found
that the effect of orthographic regularity, whereby regular words
(e.g., hint) are named more quickly than irregular words (e.g.,
pint), does not occur in the lexical-decision task. In the Waters and
Seidenberg study, this restriction was obtained when subjects were
required to respond under deadline conditions, although Hino and
Lupker obtained this result even when a deadline was not imposed.
For our purposes, however, it is sufficient to assume that the
lexical-decision task can be performed without relying on a phonological code that is developed to the point of articulation. We
draw a distinction between this form of phonological encoding,
which is necessary for explicit naming, and earlier phonological
representations that are constructed moments after the visual presentation of a word (e.g., Lukatela & Turvey, 1994; Perfetti &
Bell, 1991; Rayner, Sereno, Lesch, & Pollatsek, 1995). Thus, the
hypothesis that we test in contrasting word naming with lexical
decision is that suppression specifically affects the construction of
a phonological code for naming.
403
Experiments 1–2: Generality and Source of Word-Naming
Modulation
Experiments 1A and 1B
In one version of Experiment 1, we used a small set of noncolor
words, each appearing repeatedly both in the word-naming task
and in the color-naming task. The reason for using a small set of
items was that we did not initially know whether the suppression
effect arising from the color-naming task would be confined to
specific items seen in that task (Allport et al., 1994; Allport &
Wylie, 2000). If so, then using a small set of items would be
necessary for any effect to emerge. In the other version of Experiment 1, we tested the hypothesis that suppression applies to a
general processing mechanism independent of the specific items
used.
Method
Subjects. The subjects in Experiment 1A were 12 undergraduate students at the University of Victoria who received extra credit in an introductory psychology course for their participation in the experiment. All
subjects who participated in the experiments reported here were drawn
from this pool. No subject participated in more than one experiment. For
Experiment 1B, 24 subjects were tested.
Materials. Five 4-letter words were selected for the critical and practice trials of Experiment 1A. The words ranged in frequency from 12 to 127
occurrences per million (Kučera & Francis, 1967), with a median of 72.
The words selected had no particular association with color concepts. For
Experiment 1B, a critical set of 180 words was selected. The words ranged
from 4 to 7 letters in length, were one or two syllables, and ranged in
frequency of occurrence from 0 to 143 per million, with a median of 42.
Again, the words had no particular association with color concepts. The
critical words were arranged as six lists of 30 words. Assignment of these
lists of words to conditions in the experiment was counterbalanced across
subjects so that no word was seen more than once by a particular subject
and each word was seen equally often in each condition. Three of the six
lists were used for each subject. An additional set of 30 words with
characteristics similar to those of the critical words was selected for use as
practice items.
Procedure. Subjects in Experiment 1A were tested individually in a
quiet room using a Macintosh desktop computer with a color monitor. A
voice-activated relay was connected to the computer keyboard. Subjects
were presented with two blocks of 10 practice and 30 critical trials, with
each trial requiring the performance of two different tasks. One block was
the neutral condition and the other was the incongruent condition. The
order in which these blocks were presented was counterbalanced across
subjects. For both blocks of trials, the first task on each trial involved
naming aloud as quickly as possible a word printed in black. For the block
of neutral trials, the second task on each trial involved naming aloud the
color in which a row of five asterisks was presented. For the block of
incongruent trials, the second task involved naming aloud the color in
which a word was presented. For these two color-naming tasks, there were
five different colors, each used with equal frequency: blue, green, pink, red,
and yellow. With respect to critical trials, each word was presented six
times in the word-naming task in the neutral block, six times in the
word-naming task in the incongruent block, and six times in the colornaming task in the incongruent block. Trials in the incongruent block were
arranged so that no word appeared both for word naming and for color
naming on the same trial. For the practice trials, each of the words was
presented two times for word naming in each block and two times for color
naming in the incongruent block.
At the start of each block, subjects were informed that they would be
performing two different tasks on each trial and that the two tasks would
404
MASSON, BUB, WOODWARD, AND CHAN
always be performed in the same order. The nature of the two tasks
involved in each block of trials was explained, and subjects were instructed
to respond as quickly and as accurately as possible. Each block began
with 10 practice trials, followed by 30 critical trials.
Each trial began with the presentation of a vertically oriented rectangle
at the center of the computer monitor. The rectangle subtended a visual
angle of 22.6° vertically and 19.0° horizontally when viewed from 40 cm.
The rectangle was divided into equal-sized upper and lower boxes by a
horizontal line. The rectangle was in view for 255 ms, then a word was
presented in lowercase black letters in the center of the lower box. Stimuli
presented for both tasks were printed in 36-point Geneva font. A fivecharacter item subtended a horizontal visual angle of 5.0°. The word
remained in view until it was named by the subject, then it was erased.
After a 495-ms pause, the colored stimulus for the second task was
presented in the center of the upper box and remained in view until the
subject named its color. The experimenter pressed a key on the keyboard
to record whether an error or a voice-activated relay failure had occurred
on either task. The colored item and the rectangle were then erased, and
there was a pause of 990 ms before the next trial began automatically. A
brief rest was given at the end of the first block of trials and before the
instructions for the second block were given.
The same procedure as in Experiment 1A was used for Experiment 1B,
except that no word was presented to a subject more than once. To meet
this constraint, one of the three lists of 30 critical words seen by each
subject was used for the word-naming task in the neutral block of trials,
another was used for the word-naming task in the incongruent block, and
the third was used for the color-naming trials in the incongruent block.
Results and Discussion
Response latencies less than 100 ms or greater than 3,000 ms
were excluded from consideration. Boundaries for exclusion of
response latencies were set in this and in subsequent experiments
so that no more than 0.5% of responses were excluded (Ulrich &
Miller, 1994). No responses on the word-naming task were excluded on this basis. For the color-naming task, 1 response in
Experiment 1A (0.1%) and 2 in Experiment 1B (0.1%) were
excluded from the analysis.
For each subject, mean response latencies on trials with correct
responses were computed for word-naming and color-naming
tasks in each of the two blocks of trials. For each of the experiments reported here that involved separate blocks of neutral and
incongruent trials, we adopted the following procedure for data
analysis. First, the subject latency means and error percentages for
each task were submitted to analyses of variance (ANOVAs) with
order of block as an independent variable. If no effects involving
order of block were significant, an ANOVA collapsing across that
variable was computed and the results of that analysis are reported.
If effects of order were obtained, the results of the ANOVA
including that variable are reported. A Type I error rate of .05 was
used in all analyses reported here.
The mean response latency for each task in Experiment 1A is
shown in the upper panel of Figure 2 as a function of block type.
The data for Experiment 1B are in the lower panel of Figure 2. No
effects of block order were found for either task in either version
of the experiment, so the means shown in Figure 2 are collapsed
across that variable. For the color-naming task, the effect of block
type was significant in Experiment 1A, F(1, 11) ⫽ 50.38,
MSE ⫽ 1,163, with longer response latencies in the incongruent
block. The color-naming task, then, produced an interference effect, in which responding was slowed when color was carried by
Figure 2. Mean response latency on the word-naming and color-naming
tasks in Experiments 1A (word repetition) and 1B (no repetition) as a
function of block type. Exp. ⫽ Experiment. Error bars show 95% withinsubject confidence intervals (Loftus & Masson, 1994).
a word rather than by a row of asterisks. The effect found here is
similar to that reported by Dalrymple-Alford (1972) and Monsell
et al. (2001) for color naming of words unrelated to color concepts
versus color naming of rows of Xs or other unpronounceable
character strings. In Experiment 1B, an interference effect of
similar magnitude was found, F(1, 23) ⫽ 44.34, MSE ⫽ 3,149.
However, our primary theoretical interest lies in the results
involving the word-naming task. In Experiment 1A, the effect of
block type on word naming was significant, F(1, 11) ⫽ 24.72,
MSE ⫽ 706, whereby word-naming latencies were longer in the
incongruent block than in the neutral block. This outcome represents a clear interference effect on the first task (word naming) as
a function of the demands of the second task (color naming of
words vs. asterisks) performed on each trial. In Experiment 1B, a
similar interference effect was found, F(1, 23) ⫽ 29.56, MSE ⫽
986.
To determine whether the interference effect in the wordnaming task depends on using a small number of repeatedly
presented words (Experiment 1A) or a large set of words presented
only once each (Experiment 1B), we analyzed the word-naming
data from these two experiments in a single ANOVA. Experiment
and block type were the independent variables. This analysis
revealed a main effect of experiment, with shorter response latencies in Experiment 1A than in Experiment 1B (545 ms vs. 592 ms),
F(1, 34) ⫽ 4.63, MSE ⫽ 7,706. The latency advantage for Experiment 1 most likely is due to the benefits of repeated presentation
commonly seen in the word-naming task (e.g., MacLeod & Masson, 2000; Scarborough, Cortese, & Scarborough, 1977). There
was also a reliable interference effect, F(1, 34) ⫽ 47.58, MSE ⫽
895, that did not interact with experiment, F ⬍ 1. Thus, the
interference effect was of comparable magnitude, regardless of
whether a small number of words were presented repeatedly (54
ms) or a large number of words were presented once each (49 ms).
MODULATION OF WORD READING
In Experiment 1A, only 3 errors or voice-activated relay failures
were made in the word-naming task and only 2 such events
occurred in the color-naming task, so no analysis of these data was
carried out for either task. An analysis of mean error percentage in
Experiment 1B found no effects, Fs ⬍ 1.3, and the overall mean
error percentage, including trials spoiled because of inappropriate
or failed triggering of the voice-activated relay, was 2.0%. Only 2
subjects made word-naming errors in Experiment 1B (1 error
each), so no analysis of errors in this task was carried out.
The results of Experiment 1 revealed a clear interference effect
on the first task (word naming) as a function of the demands of the
second task (color naming of words vs. asterisks) performed on
each trial. This modulation of performance of the first task occurred even when target words were never repeated, suggesting
that the effect is not dependent on item-specific influences. We do
not wish to imply that a suitable arrangement of events (e.g., the
same word appearing for color naming then reappearing as the
next word-naming stimulus, or the use of color words as the
stimuli) would not reveal an item-specific component to the suppression effect we have obtained. Our point is that even when no
possibility exists for item specificity, we see clear evidence for
modulation of word naming as result of naming the color of a
printed word.
Experiment 2
We have assumed that the source of the modulation of word
naming seen in Experiment 1 is a form of suppression carryover
from the previous color-naming event. An alternative explanation
for the modulation of word naming is that it is a preparatory
process (perhaps a form of advance suppression) made in anticipation of the upcoming color-naming event. We can distinguish
between these possibilities by presenting two consecutive stimuli
on each task in an alternating-runs procedure (Rogers & Monsell,
1995). We continued with the task arrangement in which the first
task on each trial was word naming and the second task was color
naming. For color naming, two different color stimuli were presented in succession; for word naming, two different words were
presented in succession. Each of the stimuli was to be named
aloud. This arrangement would allow us to determine whether the
influence of naming the color of words as opposed to asterisks
differentially affects naming of the first versus the second noncolored word. If naming the first noncolored word takes substantially
less time than does naming the second one, then the modulation of
word naming can be ascribed to the effect of color naming on the
previous trial. This outcome would be expected if the act of
successfully naming the first word reduces the modulation effect
caused by the immediately preceding color-naming event. If, on
the other hand, it is the second noncolored word that is named in
a time that is substantially slower than the time needed to name the
first, we would conclude that the modulation of word naming is an
anticipatory process. This alternative outcome would be expected
if anticipatory preparation is applied to the stimulus that is temporally closest to the upcoming color-naming event.
In Experiment 2, we used a longer interval (3 s) between
successive trials, where a trial consisted of the presentation of two
word stimuli followed by two color stimuli. The reason for this
increase was to allow ample time for preparatory processes to be
completed before the onset of the first word stimulus of a trial.
405
Although we have no systematic evidence that this interval is
sufficient to completely eliminate preparatory influences, we note
that Allport and Wylie (2000) used intervals of between 1 and 2 s
for this purpose and deemed that to be an adequate range for
preparing to switch between word- and color-naming tasks.
Method
Subjects. Thirty-six subjects were tested in Experiment 2.
Materials. A set of 10 four-letter critical words was selected. Each
word had a frequency of 20 occurrences per million. The same five colors
as in Experiments 1A and 1B were used.
Procedure. The testing session began with a practice phase of 20
word-naming trials, in which each of the 10 critical words was presented
twice. In the task-switching phase, two successive stimuli were presented
for each task (i.e., two words to be named, then two colors to be named).
Both word-naming items appeared below a row of equal signs and both
color-naming stimuli appeared above (see explanation below). Word stimuli for each trial were sampled from the pool of 10 words so that no word
appeared more than once on a given trial. Two different colors were used
for the two color stimuli presented on each trial.
The display used for Experiment 2 was modified from that used in
Experiment 1. Rather than using a box to define the spatial location of
stimuli, we had a row of five equal signs presented in the center of the
computer monitor at the start of each trial. Word-naming items appeared
one line below this marker, and color-naming items appeared one line
above it. All stimuli were displayed in 12-point bold Courier font. A
five-character word subtended a horizontal visual angle of 2.4° when
viewed from 40 cm, and the vertical distance between the lower bound of
the word-naming stimuli and the upper bound of the color-naming stimuli
was 1.9°. We also changed the timing of the presentation of stimuli. Each
task-switching trial began with a 495-ms presentation of the row of equal
signs, followed by the appearance of a word printed in black below that
marker. The subject named the word, which was then erased. There was a
750-ms pause between the offset of the first word and the onset of the next
word. After the two words had been presented, the experimenter made two
key presses to classify the correctness of the responses to the two wordnaming stimuli. This procedure required approximately 500 to 1,000 ms
and varied in duration from trial to trial. Then the color-naming stimuli
appeared, were responded to, and were erased in the same manner. The
experimenter made two key presses to classify the correctness of the
color-naming responses. The delay between trials (i.e., the time between
the last of the two color-naming stimuli of one trial and the first wordnaming stimulus of the next trial) was approximately 3 s. The neutral and
incongruent blocks each began with 20 practice trials and were followed
by 40 critical trials.
Results and Discussion
Response latencies outside the range of 100 ms to 3,000 ms in
either the word-naming or the color-naming task were excluded
from analysis (0.5% and 0.2% of responses, respectively). An
additional 0.9% of word-naming responses and 0.9% of colornaming responses were excluded because of failed or inappropriate
triggering of the voice-activated relay.
The mean response latencies in the word-naming task as a
function of stimulus (first vs. second) and block (neutral vs.
incongruent) are shown in the upper panel of Figure 3. Block order
had no effect, so the data were collapsed across this variable. A
Type I error rate of .05 was used in all analyses. An ANOVA
indicated that response latencies were reliably longer for the first
of the two stimuli, F(1, 35) ⫽ 47.73, MSE ⫽ 1,409, and longer in
the incongruent block than in the neutral block, F(1, 35) ⫽ 34.37,
406
MASSON, BUB, WOODWARD, AND CHAN
first stimulus than on the second stimulus, F(1, 35) ⫽ 145.40,
MSE ⫽ 841. These two factors did not interact, F ⬍ 1.4, so the
longer response latency on the first stimulus indicates that there
was a significant switch cost that was independent of the nature of
the colored stimulus. An ANOVA of the percentage of error
revealed only a significant effect of block, with more errors in the
incongruent block than in the neutral block, F(1, 35) ⫽ 20.27,
MSE ⫽ 3.00.
The results of Experiment 2 provide strong evidence to support
the hypothesis that the modulation of word naming induced by the
requirement to name the color in which a preceding word appears
is due to the carryover of a form of suppression from the colornaming task. Had this modulation effect been due to an anticipatory process whereby subjects prepared themselves for the upcoming color-naming task by modulating word reading, we would have
expected to see a larger modulation effect on the second of the two
word-naming stimuli. Instead, the reverse result was obtained,
showing that the first of the two word-naming stimuli––which was
temporally closer to the preceding color-naming event–– bore the
brunt of the modulation effect. It appears, then, that the modulation
is stimulated by the previous color-naming episode and that subjects, to some degree, overcome that carryover effect in the course
of naming the first of the two word stimuli in the word-naming
task.
Experiments 3– 4: Accrual of Adaptive Control
Experiment 3
Figure 3. Mean response latency on the word-naming task and mean
response latency and mean percentage of error on the color-naming task in
Experiment 2 as a function of block type and stimulus. Two stimuli were
shown in succession for each task. Error bars show 95% within-subject
confidence intervals and are appropriate for assessing block-type and
stimulus effects.
MSE ⫽ 1,618. In addition, there was a reliable interaction between
these two variables, F(1, 35) ⫽ 12.38, MSE ⫽ 363. As Figure 3
indicates, the effect of block was nearly twice as large on the first
word-naming stimulus as on the second stimulus. Clearly, then, the
interfering influence of words for color naming exerted itself
primarily on the first of the two word-naming stimuli. In addition,
when only the data for the neutral block were considered, response
latencies were reliably longer on the first stimulus than on the
second stimulus, F(1, 35) ⫽ 21.18, MSE ⫽ 873, indicating that
there was a residual switch cost of 32 ms on the word-naming task,
even when a neutral stimulus carried the color on the color-naming
task. There were only 7 errors in the word-naming task, so no
statistical analysis of error data was conducted.
Color-naming performance was not affected by block order, so
the mean response latency and percentage of error data shown in
Figure 3 (middle and lower panels, respectively) were collapsed
across that variable. An ANOVA of the color-naming latencies
revealed a significant effect of block, F(1, 35) ⫽ 200.65,
MSE ⫽ 1,881, replicating the interference effect seen in the earlier
experiments. In addition, response latencies were longer on the
To this point, we have relied on blocked runs of incongruent or
neutral trials. The modulation of word naming that we have observed may occur immediately after a single color-naming episode
in which word naming must be avoided, or it may depend on a
learning process that accrues over multiple episodes. Varying
approaches to task switching instantiate both of these possibilities
to different degrees. Some researchers use runs of trials in which
subjects perform the same task then a switch to a competing task
(e.g., Allport & Wylie, 2000; Rogers & Monsell, 1995). Others
have used the method of alternating tasks, in which two competing
tasks are performed consecutively, and they contrast that condition
with one in which tasks are performed in pure blocks (e.g., Allport
et al., 1994; Jersild, 1927). If the modulation we have observed is
invoked after a single encounter in which a color carried by a word
must be named, then it may contribute to switch cost in both of
these types of task-switching paradigms.
In Experiment 3, we investigated the possibility that modulation
of word naming is instantiated by a single color-naming episode.
To do this, we presented single neutral trials (in which subjects
were to name a word, then name the color carried by asterisks) and
incongruent trials (in which subjects were to name a word, then
name the color carried by another word) in an alternating sequence. With this arrangement, we expected that suppression
arising from color naming of a word stimulus would be generated
during the color-naming event on an incongruent trial, then expressed on the word-naming event of the subsequent neutral trial.
Thus, a suppression effect would be indicated by longer wordnaming response times on neutral trials, relative to incongruent
trials––the opposite of what was seen in Experiments 1 and 2. This
pattern is expected because the slowing of word naming was
MODULATION OF WORD READING
407
shown in Experiment 2 to be a carryover effect from the colornaming task on the preceding trial. By alternating single incongruent and neutral trials in Experiment 3, we created a situation in
which the source of suppression (color naming of a word) occurs
just before the word-naming event in a neutral trial.
Whatever modulating effects on word naming occurred after a
color-naming event would not be the result of general switch costs,
because every word-naming event constituted a switch from color
naming. Rather, modulation effects would be the consequence of
the suppression of word naming on incongruent trials that took
place just prior to the task switch.
Method
Subjects. Eighteen subjects were tested.
Materials. A set of 120 critical words was selected. These words
ranged in length from four to six letters and in frequency from 47 to 660
(Mdn ⫽ 105) occurrences per million. The critical words were arranged
into three lists of 40 items each for counterbalanced assignment of lists to
conditions. Two lists were assigned for use on the incongruent trials (one
for the word-naming task and one for the color-naming task), and one list
was assigned to the neutral trials (for the word-naming task). A similar set
of 60 words was selected for use on practice trials. The five colors from the
earlier experiments were used.
Procedure. The procedure was similar to that of Experiment 1, except
that rather than presenting neutral and incongruent trials in separate blocks,
these trials were presented in an alternating sequence. Half of the subjects
began with a neutral trial (in which they read a word printed in black, then
named a color presented by a row of asterisks) that was then followed by
an incongruent trial (in which they named a word printed in black, then
named a color carried by a word), whereas the other half of the subjects
began the sequence of alternating trials with an incongruent trial. The
display layout conformed to the layout of Experiment 2. Each taskswitching trial began with a 495-ms presentation of the row of equal signs,
followed by the appearance of a word printed in black below that marker.
As soon as the subject named the word, it was erased. After a pause of 450
ms, the color-naming stimulus appeared above the marker. The subject
named the color, then both the marker and the color-naming stimulus were
erased. The experimenter then made two key presses to classify the
subject’s responses on the word- and color-naming trials. The monitor
remained blank for an additional 2.5 s after those key presses, then the next
trial began automatically.
Subjects first completed 40 practice trials consisting of 20 neutral and 20
incongruent trials in alternating sequence. Eighty critical trials followed in
which 40 neutral and 40 incongruent trials alternated.
Figure 4. Mean response latency on the word-naming and color-naming
tasks in Experiment 3 as a function of trial type (neutral vs. incongruent).
Error bars show 95% within-subject confidence intervals and are appropriate for assessing trial-type effects.
MSE ⫽ 1.25. In the word-naming task, however, there was no
effect of trial type on either the latency measure, F(1, 17) ⫽ 1.98,
MSE ⫽ 90, or the error measure, F ⬍ 1. Mean error percentages
were 0.3% and 0.4% on the neutral and incongruent trials,
respectively.
Single alternation of neutral and incongruent trials yielded no
reliable difference between these two trial types on the wordnaming task. This null result stands in marked contrast to the large
interference effect on word naming obtained in Experiments 1A,
1B, and 2 when neutral and incongruent trials were presented in
separate blocks. There are two possible explanations for this difference in outcomes. First, the alternation between single neutral
and incongruent trials may have prevented the accrual of modulation of word naming. In this account, the modulation of word
naming requires a form of learning over a series of successive
incongruent trials, where the color stimulus is a word. The alternative possibility is that the modulation effect is established after
just a single occasion of naming the color of a word, but it does not
dissipate over the course of the following neutral trial, where color
is carried by a row of asterisks. Thus the influence of modulation
would be present on every trial in Experiment 3 and would not be
detectable.
Experiment 4
Results and Discussion
Response latencies outside the range of 100 ms to 1,000 ms in
the word-naming task or outside the range of 100 to 3,500 ms in
the color-naming task were excluded from analysis (0.1%
and 0.4% of responses, respectively). Failed or inappropriate triggering of the voice-activated relay led to the exclusion of an
additional 0.5% of word-naming responses and 0.6% of colornaming responses.
Mean response latencies in the word-naming and color-naming
tasks are presented in Figure 4. A Type I error rate of .05 was used
in all analyses. For the color-naming task, response latencies were
reliably longer on incongruent trials than on neutral trials, F(1,
17) ⫽ 103.54, MSE ⫽ 707. There was also an interference effect
in the error percentage for the color-naming task, with more errors
on incongruent trials (1.6% vs. 0.3%), F(1, 17) ⫽ 7.36,
To obtain stronger evidence on the question of whether modulation of word naming accrues over multiple color-naming episodes, we conducted a further experiment in which subjects performed alternating runs of 2 neutral trials and 2 incongruent trials
or alternating runs of 10 neutral trials and 10 incongruent trials. By
comparison, Experiments 1 and 2, which showed strong modulation of word naming, presented a single block of 40 or 60 neutral
or incongruent trials (including practice and critical trials). If
modulation is absent after only a single incongruent trial, as
implied by Experiment 3, and accumulates gradually over a consecutive sequence of color-naming episodes, we would expect a
small modulation effect in word naming with runs of just 2 trials
and a much larger effect for runs of 10 trials. In both cases,
however, word naming in the neutral trials should approach preinterference levels after some number of trials.
408
MASSON, BUB, WOODWARD, AND CHAN
In addition to addressing the question of accrual of modulation,
this experiment allowed us to study three fundamental issues
regarding the nature of this modulation and its potential adaptive
significance in resolving the conflict experienced when responding
to a bivalent stimulus. First, assuming that modulation is learned,
it is possible that what is established is a mode of responding that
can be fully reinstantiated when cued by a single bivalent stimulus
event. In that case, once the modulation effect has been formed
(e.g., after the first run of 10 incongruent trials), it should reemerge
almost full-blown after the first trial of the subsequent run of
incongruent trials. Alternatively, the modulation may have to be
relearned gradually at the beginning of each run of incongruent
trials. Second, on the assumption that modulation dissipates over
the course of a series of neutral trials, we can ask whether that
dissipation occurs slowly or immediately after a single trial. Finally, we inquire into the adaptive significance of modulation. If
the modulation of word naming plays a causal role in reducing the
interference experienced during color naming of a bivalent stimulus, then when modulation of word naming is instantiated, so
should interference in color naming diminish.
Method
Subjects. Thirty-six subjects were tested. By random assignment, 12
subjects were in the runs-2 condition and 24 were in the runs-10 condition.
A larger number of subjects were tested in the runs-10 condition because
we were interested in tracking performance across trials within a run. To
obtain an adequate number of observations at each trial position within a
run, we needed more subjects in the runs-10 condition.
Materials. The critical items were 300 words that were four or five
letters in length, ranging in frequency from 6 to 967 occurrences per
million (Mdn ⫽ 40). An additional set of 30 words was created for practice
trials. The five colors from the earlier experiments were used. Assignment
of critical words to neutral and incongruent trials and to word-naming or
color-naming tasks was random. Each critical word was presented exactly
once to each subject.
Procedure. The procedure was similar to that of Experiment 3, except
that the alternating sequence of neutral and incongruent trials consisted of
runs of 2 trials of each trial type for one group of subjects (runs-2
condition) and runs of 10 trials of each type for the other group (runs-10
condition). A run of 2 or 10 trials of one type (e.g., incongruent trials) was
followed by an equally long run of trials of the other type (e.g., neutral
trials). Half of the subjects in each run condition began with a run of neutral
trials, and the other half of the subjects began with a run of incongruent
trials. Trial events were the same as in Experiment 3. For subjects in the
runs-2 condition, the first 5 runs of each trial type were treated as practice
trials, and for subjects in the runs-10 condition, the initial run of each trial
type was treated as practice. Thus, each group of subjects was provided
with 10 practice trials of each type. After the practice trials, 200 critical
trials were presented. For the runs-2 condition, the critical trials consisted
of 50 alternating runs of 2 incongruent and 2 neutral trials. For the runs-10
condition, the critical trials were presented as 10 alternating runs of 10
incongruent and 10 neutral trials.
were excluded because of artifactual or failed triggering of the
voice-activated relay. A Type I error rate of .05 was used in all
analyses.
Error rates in the word-naming task were very low, averaging 0.2% across subjects, and did not significantly vary as a
function of either run condition (2 vs. 10) or trial type (neutral vs.
incongruent). In the color-naming task, error rates did not vary
across run condition, but there were reliably more errors in the
incongruent condition than in the neutral condition (3.0%
vs. 0.3%), F(1, 34) ⫽ 35.49, MSE ⫽ 3.39.
Our analyses of response latency data were conducted separately for the runs-2 and the runs-10 conditions and emphasized
the modulation of latency both across trials within a run and
between trial types. Mean response latencies for each run condition
are shown in Figure 5. For the runs-2 condition, color-naming
performance was analyzed in an ANOVA with trial position within
a run (1 vs. 2) and trial type (neutral vs. incongruent) as the
independent variables. The only significant effect in this analysis
was trial type, F(1, 11) ⫽ 71.87, MSE ⫽ 1,525, indicating that
color-naming responses were significantly faster in the neutral
condition.
A comparable ANOVA applied to word-naming latencies in the
runs-2 condition revealed no main effects, but a there was a
significant interaction between trial position within a run and trial
type, F(1, 11) ⫽ 12.34, MSE ⫽ 139. Figure 5 shows that the source
Results and Discussion
Response latencies outside the range of 100 ms to 2,000 ms on
the word-naming task and those outside the range of 100 ms
to 3,500 ms on the color-naming task were excluded from analysis.
This constraint resulted in the removal of 0.2% and 0.3% of
word-naming and color-naming observations, respectively. A further 3.1% and 3.7% of word- and color-naming trials, respectively,
Figure 5. Mean response latency on the word-naming and color-naming
tasks in Experiment 4 as a function of block type and trial within a run.
Error bars show 95% within-subject confidence intervals and are appropriate for assessing block-type and trial effects. Error bars for the colornaming task in the runs-2 condition are smaller than the symbols.
MODULATION OF WORD READING
of this interaction was a slowing of naming responses on the first
trial of a run of neutral trials (which would have followed a run of
two incongruent trials), relative to response latency on the first trial
of a run of incongruent trials (which would have followed a run of
two neutral trials), and an elimination of this difference on the
second trial of a run. This first-trial difference of 19 ms is much
smaller than the 50-ms modulation of word naming seen in the
earlier experiments. Nevertheless, it is reliable and larger than the
nonsignificant effect seen in Experiment 3, where single neutral
and incongruent trials alternated. Thus, at least part of the modulation effect was able to develop even when subjects were given
just two consecutive trials of each type.
Moreover, naming latency quickly changed in response to the
color-naming stimulus on the first trial of a run. After that colornaming event, the second word-naming stimulus in an incongruent
run evoked a slower response, relative to the first trial, whereas in
a neutral run, the second word-naming stimulus got a response
more quickly than did the first. This pattern suggests that subjects
invoked a small modulation of word naming after the first colornaming stimulus in an incongruent run, and they quickly began to
release that modulation after receiving the first set of colored
asterisks in a neutral run. The fact that modulation occurred in the
aftermath of a color-naming event supports the conclusion drawn
from the results of Experiment 2 that modulation is a carryover
effect generated by conflict in a color-naming task rather than an
anticipatory response.
The results from the runs-10 condition yielded a pattern that was
similar in some key respects to that seen in the runs-2 condition,
although the effect sizes were much larger. In the case of word
naming, an ANOVA with trial position within a run and trial type
as the independent variables found a significant effect of trial type,
F(1, 23) ⫽ 36.55, MSE ⫽ 1,820, with slower responding on
incongruent trials. It is important to note that the trial-type effect
was reversed for the first trial position, whereby naming latency
was longer in the neutral run than in the incongruent run. As in the
case of the runs-2 condition, then, the first naming response after
a run of incongruent trials (which started a run of neutral trials)
was slower than the first naming response after a run of neutral
trials (which started a run of incongruent trials). For the runs-10
condition, this effect was 56 ms–– quite similar to that seen when
incongruent and neutral trials were fully blocked. The changing
patterns of word-naming latencies across trials within neutral
versus incongruent runs generated a significant interaction, as was
the case in the runs-2 condition, F(9, 207) ⫽ 7.76, MSE ⫽ 1,469.
Figure 5 shows that by the second trial in a run, much of the
change in naming latency had already taken place. That is, modulation was quickly invoked in an incongruent run and quickly
dissipated in a neutral run.
The difference in the magnitude of the word-naming modulation
seen in the runs-2 versus runs-10 conditions might, as we suggest,
be due to differences in the degree to which subjects can learn to
modulate word naming under these two conditions. Alternatively,
subjects might have learned modulation equally well in both
conditions but strategically deployed this modulation to different
extents. In the runs-2 condition, the frequent occurrence of task
switching might have led subjects to avoid complete instantiation
of modulation because they were aware that the task would change
soon. Assuming that invoking modulation is effortful, this strategic
avoidance would be reasonable. However, the runs-10 data show
409
that modulation can be invoked immediately after the first colornaming response in the incongruent condition and terminated after
the first encounter with a color-naming stimulus in the neutral
condition. Given the immediacy of these changes, applying and
terminating modulation may not be sufficiently effortful to motivate deliberate, strategic avoidance. Moreover, as we show below,
a cost is associated with such a strategy with respect to colornaming performance.
The color-naming data for the runs-10 condition provide a
further important fact. Figure 5 shows that in the incongruent runs,
color naming on the initial trial was much slower than on later
trials. An ANOVA found a significant effect of trial type, with
much slower responding on incongruent trials, F(1, 23) ⫽ 164.72,
MSE ⫽ 8,306. In addition, there was a significant effect of trial
position within a run, F(9, 207) ⫽ 2.15, MSE ⫽ 1,788, and the
interaction between trial type and trial position approached significance, F(9, 207) ⫽ 1.91, MSE ⫽ 1,819, p ⫽ .052. There clearly
was no variation in color-naming latency across trials within
neutral runs, F ⬍ 1, but there was a significant reduction in latency
across trials in incongruent runs, F(9, 207) ⫽ 2.49, MSE ⫽ 2,524,
with much of that reduction occurring between the first and second
trial of a run. Thus, the onset of word-naming modulation after the
first trial of an incongruent run closely coincided with a reduction
in color-naming latency.
No such reduction in color-naming latency was observed in the
runs-2 condition, indicating that the degree of word-naming modulation applied there was insufficient to benefit color naming. This
outcome reduces the appeal of the idea that subjects strategically
avoided complete modulation of word naming in the runs-2 condition, as discussed above. Avoiding complete modulation would
entail forfeiting a substantial benefit in color naming that is clearly
evident in the runs-10 condition. Thus, our preferred interpretation
of the difference in the amount of modulation of word naming
observed in the runs-10 versus runs-2 condition is that the ability
to modulate requires a relatively long sequence of incongruent
trials to develop.
One goal of Experiment 4 was to determine whether the absence
of modulation observed in Experiment 3 reflected a failure of
learning or the lack of dissipation of modulation across alternating
neutral and incongruent trials. The results of Experiment 4 provide
an unambiguous answer to this question, namely, modulation is
learned over the course of as many as 10 consecutive incongruent
trials but is weakly present even when only 2 consecutive incongruent trials are run. Therefore, it is not the case that the failure to
observe modulation in Experiment 3 reflects persistence of modulation across incongruent and neutral trials; instead, it was due to
the lack of sufficient opportunity for modulation to accrue.
Having shown that the initial learning of modulation requires a
series of consecutive trials, the question becomes whether the
reinstatement of modulation after dissipation induced by a series of
neutral trials is similarly a gradual process or can be effected
immediately after a single incongruent trial. It is clear from Figure 5, particularly from the runs-10 data, that once modulation has
been established, presumably by the initial practice run or shortly
thereafter, it is immediately reinvoked after just one trial of naming
a color carried by a word. As for dissipation of modulation, it is
equally clear that word naming returns to nearly normal speed
within a very few neutral trials after a run of incongruent trials. In
addition, it can be seen that color naming in incongruent trials
MASSON, BUB, WOODWARD, AND CHAN
410
benefits directly from modulation of word naming. As wordnaming latency increases on incongruent trials, so does colornaming latency decrease in a contingent fashion. Therefore, we
have strong evidence of an adaptive, causal relationship between
the modulation of word naming and the relative efficiency of color
naming under conditions of word– color interference. We note that
the features of this dynamic relationship include rapid reinstantiation of learned modulation and rapid recovery to baseline performance when modulation is no longer required. This pattern represents a remarkably adaptive solution to the problem of conflict
resolution when cognitive processes are in competition.
Experiments 5–7: Process Specificity of Modulation
Experiment 5
The first four experiments established three important facts
regarding the modulation effect that were revealed by our taskswitching paradigm. First, the effect is not due to repetition of
items and therefore is a general one that affects some processing
mechanism. Second, the modulation is not an anticipatory effect
but a carryover from prior color naming when the color stimulus
was a word. Third, modulation accrues over trials of the incongruent type and quickly dissipates when a run of neutral trials is
encountered; it is also quickly reinvoked once a new run of
incongruent trials begins. The next three experiments were designed to allow us to examine the question of the nature of the
word-processing mechanism involved in the modulation effect. In
Experiment 5, we tested the hypothesis that visual encoding processes in general are affected rather than some process specific to
word reading. Therefore, we examined whether modulation was
present even in a task requiring visual comparison of unfamiliar
characters that have no associated verbal label.
Results and Discussion
Response latencies in the visual-comparison task and in the
color-naming task that were outside the range of 100 ms to 2,000
ms were excluded (0.3% for both tasks). A further 0.8% of
color-naming responses were lost due to voice-activated relay
failures. A Type I error rate of .05 was used in all analyses.
Mean latencies for correct responses on the color-naming task
are shown in the lower panel of Figure 6. The data for subjects who
received the neutral block first and for subjects who received the
incongruent block first are presented separately because the blockorder variable influenced color-naming times. As the figure illustrates, mean response latency was reliably longer in the incongruent block, F(1, 14) ⫽ 102.79, MSE ⫽ 385, but this effect was
stronger when the neutral block was presented first, producing a
Block Type ⫻ Block Order interaction, F(1, 14) ⫽ 6.91, MSE ⫽
385. The main effect of block type is consistent with the results
from the earlier experiments, but it is unclear why the effect was
modulated by block order. One way of interpreting the interaction
is to note that subjects generally were slower on the second of the
two blocks of trials they received, regardless of whether it was the
neutral block or the incongruent block. It is not clear why this
effect arose in this experiment, as it did not appear in any other
experiment reported here. Only 1 error was made by 1 subject in
Method
Subjects. Sixteen subjects were tested.
Materials. A set of 8 pseudo-Chinese characters were created for use
in a visual-comparison task, and a set of 60 words was constructed for use
in the color-naming task. The normative frequency of these words ranged
from 5 to 109 occurrences per million, with a median frequency of 35.
Procedure. The same procedure as Experiment 1 was used, with the
following exceptions. First, the display characteristics were the same as in
Experiments 2– 4. Second, the testing session began with 30 practice trials
of the visual-comparison task. On each trial, two pseudo-Chinese characters appeared side by side, below the string of equal signs, until the subject
pressed one of two keys on a response box to indicate whether the
characters were the same or different. Two characters subtended a horizontal visual angle of 2.0° and a vertical visual angle of 0.9° when viewed
from 40 cm. For task-switching trials in the neutral and incongruent blocks,
the visual-comparison task replaced the word-naming task. Words used in
the color-naming task in the incongruent block were randomly assigned to
either the practice trials or the critical trials. For the practice trials in the
visual-comparison task and for each block of task-switching trials, the two
displayed characters were identical on half the trials and different on the
other half. Across the two blocks of task-switching trials, 6 of the identical
character pairs appeared eight times each and the other 2 identical pairs
appeared six times each. Of the 56 possible different pairs (taking into
account left–right position of each character), 52 appeared once each and
the remaining 4 appeared twice each.
Figure 6. Mean response latency and mean percentage of error on the
visual-comparison task and mean response latency on the color-naming
task in Experiment 5 as a function of block type and order of block type
(neutral block first [N-I], incongruent block first [I-N], and average
[Avg.]). Error bars show 95% within-subject confidence intervals and are
appropriate for assessing block-type effects.
MODULATION OF WORD READING
the color-naming task, so no analysis of color-naming errors is
reported.
The mean response latencies and mean percentages of errors on
the visual-comparison task are shown in the upper and middle
panels of Figure 6, respectively. This figure shows data in the
visual-comparison task collapsed across the variable of same versus different character pairs. We averaged across this variable
because it had no effect, nor did it interact with block type or block
order (Fs ⬍ 1). Typically, responses to same pairs are faster than
responses to different pairs in visual matching tasks, but the size of
this effect depends substantially on the extent to which different
pairs are discriminable from one another (D. A. Taylor, 1976). In
Experiment 5, the eight characters were visually very distinct, and
this is probably why there was no advantage for the same condition. An ANOVA of response latencies with block type and block
order as the independent variables found no reliable effects,
Fs ⬍ 1.4. The main effect of block type was only 3 ms, with the
larger mean in the neutral block rather than the incongruent block,
clearly indicating that there was no hint of an interference effect on
the visual-comparison task. The power of this experiment to detect
an interference effect of approximately the same magnitude as that
seen in the word-naming task of Experiment 1 (50 ms) was
estimated to be .87. An ANOVA of the error data revealed one
significant effect, an interaction between block type and block
order, F(1, 14) ⫽ 5.02, MSE ⫽ 1.86. This interaction indicates that
error rates were higher in the incongruent block when that block
came first. Overall, however, there was no main effect of block
type, F(1, 14) ⫽ 2.79, MSE ⫽ 1.86. The higher error rate in the
incongruent block was mainly due to 2 subjects who received that
block first and whose error rates were in excess of 7%.1
These results suggest that the modulation effect is not due to a
general influence of color naming of words on encoding of arbitrary visual forms. Further, the effect does not seem to operate on
general decision or response generation mechanisms involved in
the binary decision making of the sort required for the visualcomparison task.
Experiment 6
We have demonstrated that there is substantial modulation of
word-naming performance consequent to the naming of a word’s
color. This effect vanishes when a visual-comparison task is substituted for word naming, indicating that modulation is specific to
some process that is part of the word-naming task. The possibility
that we examined in Experiment 6 is that modulation affects some
stage or stages of phonological encoding of an orthographic pattern that is or are necessary for the development of an articulatory
code. As indicated above, there is evidence to suggest that the
lexical-decision task can be performed without constructing such a
code (Hino & Lupker, 2000; Waters & Seidenberg, 1985). Therefore, we used a lexical-decision task in place of the word-naming
task. If the phonological coding hypothesis is correct and if the
lexical-decision task as conducted here can be completed without
such coding, then we would expect little or no modulation of
lexical decisions by naming the colors of colored words. Although
the failure to find modulation of lexical decisions (in what is
clearly a demanding reading task) may seem counterintuitive, we
are not saying that this task would not yield switch costs if it were
used in standard task-switching paradigms. In these other situa-
411
tions, switch cost typically is based on a comparison between runs
of repeated performance of a given task and performance after a
task switch. As we have argued above, the modulation effect we
have obtained is inherently part of the switch cost if switching
occurs between competing tasks applied to bivalent stimuli. However, our paradigm measures the modulation of a task with other
sources of switch cost removed because those sources are present
in both incongruent and neutral conditions.
Method
Subjects. Eighteen subjects were tested.
Materials. A set of 60 critical words and 60 critical pronounceable
nonwords was constructed. These items were all four to six letters in
length. The normative frequency of the words ranged from 34 to 787
occurrences per million, with a median of 125. The 60 critical words
and 60 nonwords were each arranged as three lists of 20 items for
assignment to tasks. One list of each type of item was assigned to serve as
letter strings on the lexical-decision task in the neutral block, another set of
word and nonword lists was assigned to serve as letter strings on the
lexical-decision task in the incongruent block, and the remaining lists were
to serve as color stimuli on color-naming trials in the incongruent block.
Assignment of the lists of words and nonwords to these three tasks was
counterbalanced across subjects. An additional set of 20 words and 20
nonwords was created for use on lexical-decision practice trials, and a set
of 30 words and 30 nonwords was constructed for use on task-switching
practice trials. The same five colors as in the previous experiments were
used for the color-naming task.
Procedure. The procedure was identical to that of Experiment 5,
except that the first task on every task-switching trial required subjects to
classify a letter string printed in black as a word or a pronounceable
nonword. For this lexical-decision task, the stimulus was equally often a
word or a nonword. As before, the stimulus presented for the second task
was printed in color, and the task was to name that color. In the incongruent
block, the stimulus that carried the color was equally often a word or a
pronounceable nonword. In the neutral block, the color was always carried
by a row of asterisks. Subjects pressed one of two labeled buttons mounted
on a response box as rapidly as possible to indicate their decision. The
testing session began with 40 practice trials involving just the lexicaldecision task. The practice phase was followed by a neutral block and an
incongruent block of task-switching trials, with block order counterbalanced across subjects. Each task-switching block began with 20 practice
trials followed by 40 critical trials.
Results and Discussion
Response latencies outside the range of 100 ms to 3,000 ms on
the color-naming task (0.4%) and response latencies outside the
range of 100 ms to 2,000 ms in the lexical-decision task (0.2%)
were excluded from the analyses reported here. In addition, 2.1%
of color-naming trials were excluded because of inappropriate
triggering or failures of the voice-activated relay.
Block order had no effect on color-naming performance, so we
present data from that task collapsed across that variable. A Type
I error rate of .05 was used in all analyses. Two analyses were of
1
To check the possibility that the influence of block type on error
reflects a speed–accuracy trade-off that prevented an interference effect
from emerging in the latency data, we reanalyzed the data while excluding
the 2 subjects whose error rates were over 7% in the incongruent block.
With those 2 subjects excluded, there was no sign of an influence of block
type on response latency or error rate, Fs ⬍ 1.
412
MASSON, BUB, WOODWARD, AND CHAN
interest. In the first, we compared mean latency of color-naming
responses for word and nonword stimuli in the incongruent block.
Mean latency was reliably shorter for words (654 ms) than for
nonwords (683 ms), F(1, 17) ⫽ 6.86, MSE ⫽ 1,049. In the second
analysis, we compared color naming in the neutral and incongruent
blocks, averaging across words and nonwords in the latter case. As
in the earlier experiments, latencies were longer in the incongruent
block (668 ms vs. 592 ms), F(1, 17) ⫽ 67.29, MSE ⫽ 781. Four
subjects made 1 error each in the incongruent block, and no errors
were made in the neutral block. Because there were so few errors,
no statistical analysis of error data was conducted.
The color-naming data, then, produced one new result, namely,
that responses were slower to nonword items than to words. This
outcome is similar to the small color-naming latency advantage
found by Monsell et al. (2001) for high-frequency words over
pseudowords. Most of the words used in the present experiment
were in the same frequency range as Monsell et al.’s highfrequency words, so our result is consistent with their finding. In
addition, however, the pseudowords used by Monsell et al. were
matched to the words on a variety of variables, such as length and
onset. Our nonwords were not matched to the words. For example,
whereas 12 of the critical words shared onsets with one of the five
color names used in the experiment, 20 of the nonwords did so.
The difference in color-naming latency found here, then, could be
due in part to differences between the words and nonwords on
some structural characteristic, or it could be due to the context of
task switching in which the color-naming task was embedded.
Further work is needed to examine these possibilities.
Mean response latencies and the percentage of error for word
and nonword targets on the lexical-decision task are shown in
Figure 7. Separate analyses were conducted for each type of item,
although it is clear from Figure 7 that responses to words were
made more quickly and more accurately than were responses to
nonwords. We applied an ANOVA to latency data for word targets
and found no effect of block type or block order, Fs ⬍ 1, although
the interaction approached significance, F(1, 16) ⫽ 4.28,
MSE ⫽ 1,075. The pattern of means for the interaction indicates
that subjects were about 20 ms slower in the second block of trials,
regardless of whether that block was the neutral or incongruent
block. There clearly was no evidence of an interference effect on
lexical-decision trials involving words, as the mean latencies for
the neutral and incongruent blocks were nearly identical (622 ms
vs. 623 ms).
An ANOVA of the nonword latency data revealed a significant
interaction, F(1, 16) ⫽ 4.53, MSE ⫽ 3,168, with the opposite
pattern to that seen with words. Namely, there was a reduction in
response latency of about 40 ms on the second block of trials.
Neither the block-type nor the block-order main effects were
reliable, Fs ⬍ 1. As with word targets, there was no sign of an
effect of block type (776 ms vs. 772 ms for the neutral and
incongruent blocks, respectively). The interaction effects seen with
word targets and with nonword targets indicate that subjects
slowed their responses to words and speeded their responses to
nonwords as they moved from the first to the second block of
trials, an effect reminiscent of the criterion shift reported for word
naming by T. E. Taylor and Lupker (2001). However, this pattern
was unaffected by whether asterisks or letter strings carried the
color on the color-naming task. Analyses of the percentage of error
Figure 7. Mean response latency (upper panels) and percentage of error
(lower panels) for words and for nonwords on the lexical-decision task in
Experiment 6 as a function of block type and order of block type (neutral
block first [N-I], incongruent block first [I-N], and average [Avg.]). Error
bars show 95% within-subject confidence intervals and are appropriate for
assessing block-type effects.
produced no significant effects, Fs ⬍ 1 for words and Fs ⬍ 2.6 for
nonwords.
The primary result of interest in Experiment 6 was the lack of an
influence of block type on lexical-decision responses either for
words or for nonwords. When these two item types were combined, there still was no effect of block type, F ⬍ 1. Because of the
relatively greater stability of response latencies to word targets, it
was in this condition that Experiment 6 had the greatest power to
detect an effect of block type on lexical-decision responses. The
estimated power to detect an effect of the magnitude seen in
Experiment 1 (50 ms) was .99.
The lack of a modulation effect on lexical decision is consistent
with the idea that an aspect of phonological encoding, used in word
naming but apparently not in lexical decisions, is at least one of the
processes affected by modulation. Although we can tentatively
conclude that a locus of modulation is specific to the construction
of a phonological code, another possibility exists. In substituting
the lexical-decision task and, for that matter, the visualcomparison task of Experiment 5 for the word-naming task, subjects responded by making a binary decision and a manual response instead of a vocal one. If the modulation effect seen in the
word-naming experiments is simply attributable to suppression of
an overt articulatory response, then no such effect should be seen
when a task is used that still requires construction of a phonological code but that no longer demands overt articulation.
Experiment 7
To rule out the possibility that modulation effects emerge as the
result of the suppression of an overt articulatory response, we used
MODULATION OF WORD READING
a word-reading task that does not require subjects to overtly
articulate but continues to demand the construction of a phonological code. The task we used was a phoneme-detection task in
which subjects were to decide whether a target word contained a
specified phoneme. Responses were made manually, as in Experiments 5 and 6, and required a binary decision (i.e., was the
phoneme present or absent).
Method
Subjects. Twenty-four subjects were tested.
Materials. A set of 120 critical words, all containing one occurrence of
the letter a, was constructed. For 60 of these items, the a contributed to
producing a long a sound (/e/) in the normal pronunciation of the word
(e.g., wave, drain, stay), whereas for the other 60 words, the a participated
in producing a different sound (e.g., share, dream, cattle). The critical
words ranged from four to six letters in length and from 16 to 808
occurrences per million in frequency, with a median frequency of 97.
The 60 critical words of each type were arranged as three lists of 20 items
each for assignment to the three different conditions of the experiment:
phoneme detection in the neutral block, phoneme detection in the incongruent block, and color naming in the incongruent block. Assignment of
lists to conditions was counterbalanced across subjects so that each word
was seen exactly once by each subject and each word appeared equally
often in each condition. Additional sets of 40 and 60 words, each with half
of its words containing the critical /e/ sound, were constructed for use in a
phoneme-detection practice phase of the experiment and for practice on
task-switching trials, respectively. For the color-naming task, the same
colors were used as in the previous experiments.
Procedure. The procedure was identical to that of Experiment 5,
except that subjects were instructed to determine whether the designated
phoneme /e/ was present in a word instead of being asked to make a lexical
decision. /e/ words and non-/e/ words appeared with equal frequency in
both phoneme-detection and color-naming tasks. Subjects responded by
making manual responses, as in Experiments 5 and 6.
413
in the analyses we report. We simply note that subjects were
quicker to respond when the /e/ sound was present in a word than
when it was absent (901 ms vs. 1011 ms), F(1, 22) ⫽ 36.98,
MSE ⫽ 7,858. Mean response latencies and the percentage of error
for the phoneme-detection task are shown in Figure 8. An ANOVA
of the latency data revealed a significant effect of block, F(1,
22) ⫽ 4.92, MSE ⫽ 3,580, with longer mean latency in the
incongruent block than in the neutral block (976 ms vs. 937 ms).
This 39-ms effect is of comparable magnitude to that seen in
Experiments 1 and 4. Although the effect of block order was not
significant, F ⬍ 1, that variable did interact with block type, F(1,
22) ⫽ 7.88, MSE ⫽ 3,580.
As can be seen in Figure 8, the interaction arose because the
block-type effect was present only among subjects who were given
the incongruent block first. This pattern of results reflects a practice effect, whereby subjects responded more quickly on the second block of trials than on the first block of trials in the phonemedetection task. The lack of practice when the neutral block was
presented first obscured the interference effect for that group of
subjects, just as it exaggerated the effect for subjects who were
presented the incongruent block first. Averaging over these two
groups of subjects takes into account the influence of practice and
yields a significant main effect of block type, as reported above.
Had there been no underlying effect of block type and only an
effect of practice, we would have obtained a clear crossover
interaction like that seen for lexical decision in Figure 7. The
ANOVA applied to the percentage of error data for the phonemedetection task produced no reliable effects.
The results of Experiment 7 clearly demonstrate that modulation
of word processing can be found in a task that requires a binary
decision and manual responding instead of a naming response.
Moreover, we conducted an additional task-switching experiment,
not reported in detail here, that was similar to Experiment 6, except
Results and Discussion
Response latencies outside the range of 100 ms to 3,000 ms on
the color-naming task (0.3%) and on the phoneme-detection task
(0.1%) were excluded from analyses. In the color-naming task, an
additional 2.4% of trials were excluded because of inappropriate
activation or failures of the voice-activated relay.
Color-naming response latencies were not affected by block
order, so the analyses we report here collapsed over that variable.
A Type I error rate of .05 was used in all analyses. Color naming
in the incongruent block was not reliably different for words
containing the /e/ sound and for words not containing that sound
(686 ms vs. 696 ms), F ⬍ 1. However, mean response latency was
significantly longer in the incongruent block than in the neutral
block (691 ms vs. 592 ms), F(1, 23) ⫽ 110.81, MSE ⫽ 1,058.
Across all subjects, a total of 2 color-naming errors were made in
the neutral block and 3 errors were made in the incongruent block,
so no statistical analyses of errors were carried out. These results
show that, as in the earlier experiments, subjects took substantially
longer to name colors when they were embodied in words (or letter
strings) than when they were carried by a row of asterisks.
Response latencies on the phoneme-detection task were modulated by block order, so the analyses reported here included that
variable. The presence versus absence of the target /e/ sound did
not interact with any factors, so we collapsed across that variable
Figure 8. Mean response latency on the phoneme-detection task in Experiment 7 as a function of block type and order of block type (neutral
block first [N-I], incongruent block first [I-N], and average [Avg.]). Error
bars show 95% within-subject confidence intervals and are appropriate for
assessing block-type effects.
MASSON, BUB, WOODWARD, AND CHAN
414
that instead of presenting words and nonwords in a lexicaldecision task, we presented pseudohomophones (e.g., brane, hoap)
and pronounceable nonwords (e.g., naft, perd). Subjects classified
items using manual responses according to whether the items
would sound like real words if pronounced aloud (a phonological
lexical-decision task). As in the phoneme-detection task of Experiment 7, subjects showed a significant interference effect of 41 ms
on this phonological lexical-decision task such that longer decision
times were observed in the incongruent block (where the color in
the color-naming task was carried by a pseudohomophone or a
nonword) than in the neutral block (where the color was carried by
asterisks). Furthermore, we also conducted a replication of Experiment 7 and obtained a significant 63-ms interference effect on the
phoneme-detection task. Thus, when the task that precedes color
naming requires phonological processing, response latency is reliably modulated by the nature of the stimulus presented in the
color-naming task. We found no modulation in a visualcomparison task that did not require word reading, nor did we find
modulation in a lexical-decision task. It remains to be seen whether
modulation can be found in other tasks, especially reading tasks,
that do not rely as heavily on phonological processes as do word
naming or phonological decisions.
General Discussion
The literature on the cost of switching between tasks includes
accounts based on endogenous shifting of task set (e.g., Baddeley et al., 2001) and exogenous cuing of task set (e.g., Rogers
& Monsell, 1995). An additional claim holds that shifting from
a bivalent stimulus incurs a persistent cost associated with
resolving the competition involved in making a nondominant
response to one dimension of the stimulus and avoiding a
dominant response to the other dimension (Allport et al., 1994;
Allport & Wylie, 2000). In typical task-switching paradigms
that use bivalent stimuli, there is the possibility that the persistent cost of competition resolution is a substantial component of
the observed task-switch cost. It should be possible to observe
the cost of this competition resolution on subsequent performance of the dominant task even when it is applied to a
univalent stimulus. The present experiments were designed to
investigate this persistent cost, separate from the influences of
endogenous and exogenous control of task set. Both the neutral
trials and the incongruent trials used in our experiments involved a switch between word naming and color naming; therefore, both involved a switch cost of some form. The crucial
difference between these two trial types was that in the incongruent trials, color naming was performed on a bivalent stimulus (a colored word), whereas in the neutral trials, color
naming was applied to a univalent stimulus (a colored string of
asterisks). Thus, competition resolution would arise only in the
incongruent trials. The persistent cost of this resolution was
revealed by comparing word-naming performance (in a task
where the words were always presented in black) in incongruent
and neutral trials.
Our experiments indicate that word naming incurs a large cost
due to interference generated from naming the color in which a
previous word was presented. Strikingly, this effect was no greater
for a small set of words presented repeatedly, even in the colornaming task, than for a much larger set of words, each of which
was presented only once. Second, we showed that modulation of
word naming accrues over time, requiring between 2 and 10
consecutive incongruent trials to reach asymptotic levels. It is not
clear in exactly which run of the 10 trials modulation was learned,
but this question was not of great concern here. Instead, it is
important to note that virtually no modulation was obtained when
single incongruent and neutral trials were alternated, and only a
small amount of modulation was found when runs of 2 consecutive
trials of a particular type were presented. However, runs of 10
trials yielded a full modulation effect. Moreover, once modulation
was learned, it dissipated rapidly over the course of a few trials
when subjects shifted to a run of neutral trials. When subjects
encountered a new run of incongruent trials, the modulation effect
was reinstantiated after only one trial. Also, this reinstatement of
modulation coincided with a reduction in the latency to name the
color of words. No such change in color naming was observed in
the neutral condition (where color was carried by a row of asterisks). Thus, there is good evidence that modulation of word reading mitigates the competition experienced during the naming of a
color carried by a word.
Finally, we provided preliminary evidence that modulation of
word-identification processes was specific to tasks involving the
construction of a phonological code from a visually presented
word. In particular, although word naming and phoneme detection
showed reliable modulation, two tasks that presumably did not
require phonological encoding, visual comparison and lexical decision, did not. Our tentative conclusion, then, is that resolving
competition during the naming of word color affects processes
specific to the construction of phonological codes. Moreover, the
results of Experiment 1 imply that this influence is not confined to
the particular word that carries the named color but generalizes
across a large set of words.
Although we have shown that the modulation effect on word
naming generalizes beyond the specific words encountered in the
color-naming task, a question for future research concerns the
particular conditions of conflict resolution that lead to modulation.
In our experiments, conflict consistently involved competition
between the habitual naming response to a word and the less
practiced task of color naming. It remains to be seen whether
modulation also can be observed when neither member of a
conflicting task pair is habitual and conflict arises only because of
task-set activation. If there is nothing special about conflict based
on task dominance (i.e., one of two task sets is habitual), then we
would expect modulation to occur in any situation in which competition occurs between two active task sets. A further question
concerns the generalizability of our account to tasks other than
word reading. For example, would modulation be observed in
other naming tasks, such as picture naming or digit naming, or in
tasks not requiring responses based on the pronunciation of words?
Assuming that modulation does extend to a wider variety of task
combinations, it becomes an important and general component of
task-switch costs, as we discuss below.
Implications for Accounts of Task Switching
Many experiments on task switching indicate that costs persist
even when subjects have ample time to prepare for an upcoming
task switch (e.g., Allport & Wylie, 2000; Meiran, 1996; Rogers &
Monsell, 1995). One explanation for this residual switch cost is
MODULATION OF WORD READING
that subjects cannot fully configure a task set until it is exogenously cued by a specific stimulus (Rogers & Monsell, 1995).
Our experiments used intervals of approximately 1 s to 3 s between
response to the color stimulus and the onset of the word stimulus
on the next trial. Given previously reported results, intervals of this
length should have provided ample time for subjects to prepare for
execution of our word-identification tasks (word naming in Experiments 1– 4, lexical decision in Experiment 6, and phoneme
detection in Experiment 7), and therefore any cost of switching
from color naming to word identification can reasonably be considered residual in nature.
This observation raises the possibility that performance on our
word-identification tasks could have been determined in part by a
failure to adequately configure the task set until the occurrence of
a stimulus that acts as the exogenous cue. If we further assume that
the influence of exogenous cuing is stronger when switching from
a task involving a bivalent stimulus (naming the color of a word in
our case) than when switching from a task involving a univalent
stimulus (naming the color of a row of asterisks), then one might
argue that our modulation effect is, in fact, another instance of
residual switch cost driven by exogenous cuing. Indeed, the pattern
of switch costs seen in Experiment 2 (see Figure 3), in which we
used a procedure similar to that used by Rogers and Monsell
(1995), is consistent with this assumption. In that experiment,
switch cost, defined by Rogers and Monsell as the difference in
response latency between the first and second items presented for
word naming, was larger in the incongruent condition (involving a
switch from a bivalent stimulus) than in the neutral condition
(involving a switch from a univalent stimulus). Conceivably, the
cost of switching from naming the color of a bivalent stimulus to
naming a univalent word reflects the inability to fully configure the
word-naming task set until the occurrence of the target word. In the
case of switching from a univalent stimulus in the color-naming
task, we might conjecture that there is less dependence on exogenous cuing.
However, our experiments provide strong evidence against this
proposition. First, the need for exogenous cuing to fully establish
a task set should affect performance regardless of the particular
task to which one is switching. But we have shown that the
modulation that underlies our results—namely, the effect of bivalent color naming on a subsequent task— depends crucially on the
nature of that task. In particular, we found no evidence of modulation of either lexical-decision or visual-comparison tasks. For
both of these tasks, as with the word-naming and phonemedetection tasks, exogenous cuing ought to be necessary for task-set
configuration after the execution of bivalent color naming.
Second, the modulation effect accrues over time. It is much
greater when subjects carry out at least 10 consecutive incongruent
trials (each trial consisting of the naming of a univalent word
stimulus followed by the naming of the color of a bivalent item)
than when only 1 or 2 such trials are presented consecutively. This
observation, coupled with the task specificity discussed above,
implies that subjects are learning to modulate specific aspects of
their word-reading behavior. There is nothing in the Rogers and
Monsell (1995) account of exogenous cuing to imply that reliance
on such cuing to configure task sets is a learned skill.
Third, and most important, we have good evidence that modulation of word naming benefits performance in the color-naming
task, but only if color is carried by a word. This outcome indicates
415
that modulating a component process of word naming has adaptive
significance in a situation where word naming interferes with the
competing task of color naming. The theoretical construct of
exogenous cuing may apply to word naming in our paradigm, but
it surely has no explanatory force in regard to the link between
slowed word naming and the subsequent naming of a color carried
by a word. Modulation of word naming, once learned, results in an
immediate benefit to color naming after the first trial of a block of
incongruent trials and persists over the subsequent trials of that
block (see Figure 5). This pattern is consistent with the claim that
slowed word naming involves suppression of a component process
of reading and leads to reduced interference when naming the
color of a bivalent stimulus. There is nothing in the exogenous
cuing account that would explain why a slower word-naming
response is associated with a faster color-naming response.
We have argued that the concept of exogenous cuing cannot
provide an adequate account of the modulation effect that we have
obtained. In addition, we can consider whether some of the residual switch cost reported by Rogers and Monsell (1995) and ascribed by them to exogenous cuing might receive an alternative
interpretation based on the modulation of component processes.
The general paradigm used by Rogers and Monsell involves presenting two consecutive trials of the same task followed by another
two trials of another task, with both tasks involving bivalent
stimuli. Their definition of a switch cost is the difference between
performance on the first trial of each task and the subsequent trial.
Performance is much slower on the first trial because, in their
account, configuration of task set is incomplete until the occurrence of a relevant stimulus but is fully established by the time the
next stimulus is presented. Alternatively, the observed difference
between the first and second trial performances is fully compatible
with the idea that performance of the first trial is slow because
modulation of a component process of the relevant task has occurred as a result of interference generated during the immediately
preceding task. The modulation rapidly dissipates after a single
trial, just as it did in our Experiment 4 (see Figure 5). Thus, the
difference between two consecutive trials of the same task can be
attributed to the dissipation of modulation rather than to the
exogenous cuing of a task set (for a similar idea, see Allport et al.,
1994; Meiran, 1996; Meiran et al., 2000). We note that switch
costs found by Rogers and Monsell were generally much larger
than the modulation effects on word naming that we report. In
most of their experiments, however, both tasks involved bivalent
stimuli, whereas our word-naming task always involved a univalent stimulus. In cases in which Rogers and Monsell’s subjects
switched to a task involving a univalent stimulus (their nocrosstalk condition), switch costs were of a magnitude similar to
ours.
Another interpretation of residual switch costs is based on
transfer or priming effects from a previously executed task that
interfere with attempts to execute a currently designated task
(Allport et al., 1994; Allport & Wylie, 2000). These influences can
be of two types. First, there is persisting activation of the previous
but now conflicting response to an attribute of a particular stimulus
(e.g., having earlier named the color red carried by a word, one
must now name a word printed in red). Second, this persisting
activation could, in principle, also maintain a task set that conflicts
with performance after a task switch (e.g., having earlier performed a color-naming task in which color was carried by a word,
416
MASSON, BUB, WOODWARD, AND CHAN
one must now name a word). Both of these forms of activation
have some plausibility if the word-naming task involves presenting
a word printed in color.
The question is whether these ideas are plausible when applied
to our procedure. The stimulus-specific version of persisting activation could not operate during our experiments because the
word-naming task was always to name a word printed in black, and
black never appeared as a color in the color-naming task. It seems
unlikely that there is a tendency on the part of the subject to name
the color (black) of the printed word in the word-naming task when
no such response (“black”) was ever required in the color-naming
task. Moreover, words that appeared in the word-naming task
never were color names, so there is no reason to expect the
word-naming stimulus to evoke competing color-naming responses. Activation as the persistence of a more general colornaming procedure would have to have the following characteristics, given our results: (a) It would have to persist even though the
word-naming stimulus is univalent, (b) it would have to be specific
to tasks that appear to depend on phonological processes, and (c)
it would not be an automatic consequence of executing a single
color-naming response to a bivalent stimulus but would require
multiple trials to develop. In addition, such activation would have
to survive despite the possible effects of backward inhibition that
would mitigate against persistence of the abandoned task set of
color naming (Mayr & Keele, 2000).
Allport and colleagues (e.g., Allport & Wylie, 2000) also suggested that residual switch cost can be generated by the persistent
suppression of a response or process that competes with the
execution of a task (e.g., suppression of word naming when naming the color of a printed word). This cost may then be observed
when the subject switches to a task that requires the suppressed
response or process. A substantial contribution to switch cost of
the sort seen by Allport and colleagues is attributable to response
conflict associated with specific items. For example, naming the
color of a particular color word leads to difficulty in naming that
word on a subsequent trial because the act of reading the word has
been suppressed during the generation of the earlier color-naming
response (Allport & Wylie, 2000, Experiment 3). In addition to
their work on stimulus-specific suppression, Allport and Wylie
obtained some evidence that suppression affects a general wordnaming process by using color words never seen in their Stroop
color-naming task (although those corresponding colors had been
presented). They found that both these words and color words that
had been presented in the Stroop color-naming task showed the
same degree of slowing when presented subsequently in black for
word naming. This effect was larger than the 50-ms effect we
obtained. It is possible, then, that part of word-naming suppression
is specific to the particular semantic category, in this case colors,
that constituted the conflicting response dimension. The remaining
component appears to be quite general, extending equally to noncolor words seen either repeatedly or only once in the course of an
experimental session. Our results indicate that suppression of word
naming is even more general than suggested by Allport and colleagues’ results, not restricted either to color words or to a small
set of noncolor words repeatedly presented in the context of color
naming and word naming.
The account of task modulation that we have developed bears
some resemblance to the concept of backward inhibition, in which
a just-executed task set is inhibited to prevent it from intruding
during execution of a subsequent task (Mayr & Keele, 2000). Both
our account and the backward inhibition account assume a form of
suppression that generalizes across items. Backward inhibition is
assumed to occur unconditionally as a result of task disengagement
(Mayr & Keele, 2000, p. 23). Presumably such inhibition would
occur even when the disengaged task applies to a univalent stimulus, provided there is a possibility of upcoming competition
involving a bivalent stimulus in which the disengaged task must be
avoided (Mayr, 2001). The incongruent trials of our paradigm are
an instance of exactly this situation (naming a univalent stimulus,
then naming the color of a bivalent stimulus). However, the
modulation effect that we observed was not found under all circumstances and, in particular, did not occur when incongruent and
neutral trials were presented in alternating runs (our Experiment
3). If backward inhibition were responsible for the modulation of
word reading, disengagement of word reading on an incongruent
trial should have led to slower word reading on neutral trials.
Modulation occurred only when runs of at least two consecutive
incongruent trials alternated with runs of at least two neutral trials.
Therefore, there is reason to argue that backward inhibition is not
the mechanism responsible for the modulation of word reading we
have shown.
Learned Modulation of Component Processes
In our account, the basis for modulation of word naming is the
competition occurring during performance of the color-naming
task in the incongruent condition. Competition between task sets in
the Stroop paradigm has been modeled by Cohen, Dunbar, and
McClelland (1990), who used the concept of task nodes to represent the demands of word reading and color naming. To carry out
color naming, subjects must suppress or gate the word-reading task
node to prevent competition with the color-naming response. This
gating is only partially successful because of the highly overlearned nature of the word-reading response, yielding the interference seen when color naming occurs in the presence of a word.
What aspect of a Stroop stimulus activates the word-reading task
node? Monsell et al. (2001) suggested that it is the properties of the
letter string’s orthography, including pronounceability detected by
rapid activation of orthographic-to-phonological connections. According to Monsell et al., gating occurs at an early stage of
processing, reducing the competition between the word-reading
and color-naming task sets and, in most circumstances, preventing
the development of response competition.
There are a number of reasons to argue that this gating account
of the mechanism of competition resolution in the Stroop paradigm
is incomplete. First, contrary to the assumption made by Monsell
et al. (2001) that suppression of the word-reading task set prevents
lexical access, there is evidence that words carrying colors in a
color-naming task show priming on a subsequent lexical-decision
task equivalent to that shown by words that are read during a study
phase (Szymanski & MacLeod, 1996). However, it may be the
case that gating of early stages of reading merely slows lexical
access instead of preventing it completely, and this form of slowing is sufficient to overcome competition from the word during
color naming. But even if gating occurs at an early stage, there is
clearly a second kind of suppression that can operate in the Stroop
paradigm. We have shown that a few seconds after a color-naming
response to a colored word, no slowing of a lexical-decision task
MODULATION OF WORD READING
occurred. Thus, there appears to be no residual cost of competition
during color naming on the general task of word reading despite
the fact that substantial costs are found when word-naming responses are measured. Gating at an early stage of word reading, if
it has occurred during the color-naming task, is therefore unlikely
to exert a persistent influence on a subsequent word-reading task.
The kind of suppression that we suggest is measured in our
experiments deals with later stages of processing, where phonological codes are constructed (see Burt, 2002, for a similar proposal regarding the modulation of phonological processes in the
color Stroop task). Thus, tasks such as word naming and phoneme
detection but not necessarily lexical decision are affected. The
nature of the conditions that lead to modulation of earlier stages of
processing remains an open question. For example, presenting
words for reading in color instead of in black might invoke a form
of interference during word reading that would lead to the gating
of perceptual inputs that could have consequences on subsequent
color-naming and/or word-reading trials. For an explicit distinction
between control mechanisms differentially affecting stimulus and
response representations, see Meiran (2000).
It is interesting to consider the possible neural correlates of the
form of modulation that we have demonstrated. We know that
selective attention to particular visual dimensions such as color or
motion demonstrates enhancement of activity in domain-specific
cortical regions (Corbetta, Miezin, Dobmeyer, Shulman, & Petersen, 1990). The enhancement of activation in cortical regions
associated with particular types of knowledge can be observed
even when selective attention is applied to a bivalent Stroop-like
stimulus in which one dimension dominates another. Banich et al.
(2001) showed increased activation in a number of cortical regions
relevant to word processing by comparing brain activity during
color naming when color was carried by an incongruent color word
versus when color was carried by a neutral word. This increase in
activation was attributed to priming of the pathway between words
and color concepts as a by-product of the task demands of selectively attending to the color dimension. However, the subtraction
for regions of activation in the incongruent versus neutral conditions of the Banich et al. study is not very likely to identify the
neural mechanism that permits selective responding to a color
rather than a word when presented with a Stroop stimulus. Because
the subtraction was between two conditions in which color was
carried by a word, the resulting differences were unable to reveal
a suppression mechanism that we assume is common to both types
of words.
Our paradigm is a viable candidate for investigating the neural
mechanisms that support selective responding to a bivalent stimulus even when the word does not refer to a color. We have
demonstrated that (a) suppression applies to all words, not just
those that have appeared on color-naming trials; (b) the amount of
suppression is substantial in relation to the relatively fast and
overlearned activity of naming a word; (c) suppression may be
selective to processes specific to phonological encoding; and (d)
suppression is relatively long-lasting in that it can be detected even
after a delay of about 3 s after a color-naming response. These
characteristics invite further examination of whether it can be
shown that the activity of specific cortical regions is altered during
modulation of word naming.
417
References
Allport, A., Styles, E. A., & Hsieh, S. (1994). Shifting intentional set:
Exploring the dynamic control of tasks. In C. Umilta & M. Moscovitch
(Eds.), Attention and performance XV: Conscious and nonconscious
information processing (pp. 421– 452). Cambridge, MA: MIT Press.
Allport, A., & Wylie, G. (2000). Task switching, stimulus response bindings, and negative priming. In S. Monsell & J. Driver (Eds.), Attention
and performance XVIII: Control of cognitive processes (pp. 35–70).
Cambridge, MA: MIT Press.
Baddeley, A., Chincotta, D., & Adlam, A. (2001). Working memory and
the control of action: Evidence from task switching. Journal of Experimental Psychology: General, 130, 641– 657.
Banich, M. T., Milham, M. P., Jacobson, B. L., Webb, A., Wszalek, T.,
Cohen, N. J., & Kramer, A. F. (2001). Attentional selection and the
processing of task-irrelevant information: Insights from fMRI examinations of the Stroop task. In C. Casanova & M. Ptito (Eds.), Progress in
brain research (Vol. 134, pp. 459 – 470). Amsterdam: Elsevier.
Burt, J. S. (2002). Why do non-color words interfere with color naming?
Journal of Experimental Psychology: Human Perception and Performance, 28, 1019 –1038.
Cohen, J. D., Dunbar, K., & McClelland, J. L. (1990). On the control of
automatic processes: A parallel distributed processing account of the
Stroop effect. Psychological Review, 97, 332–361.
Corbetta, M., Miezin, F. M., Dobmeyer, S., Shulman, G. L., & Petersen,
S. E. (1990, June 22). Attentional modulation of neural processing of
shape, color, and velocity in humans. Science, 248, 1556 –1559.
Dalrymple-Alford, E. C. (1972). Associative facilitation and interference in
the Stroop color-word task. Perception & Psychophysics, 11, 274 –276.
Fagot, C. (1994). Chronometric investigations of task switching. Unpublished doctoral dissertation, University of California, San Diego.
Fox, L. A., Schor, R. E., & Steinman, R. J. (1971). Semantic gradients and
interference in naming color, spatial direction, and numerosity. Journal
of Experimental Psychology, 91, 59 – 65.
Gilbert, S. J., & Shallice, T. (2002). Task switching: A PDP model.
Cognitive Psychology, 44, 297–337.
Hino, Y., & Lupker, S. J. (2000). Effects of word frequency and spellingto-sound regularity in naming with and without preceding lexical decision. Journal of Experimental Psychology: Human Perception and Performance, 26, 166 –183.
Jacoby, L. L. (1991). A process dissociation framework: Separating automatic from intentional uses of memory. Journal of Memory and Language, 30, 513–541.
Jersild, A. T. (1927). Mental set and shift. Archives of Psychology,
14(Whole No. 89).
Klein, G. S. (1964). Semantic power measured through the interference of
words with color-naming. American Journal of Psychology, 77, 576 –
588.
Kučera, J., & Francis, W. N. (1967). Computational analysis of presentday American English. Providence, RI: Brown University Press.
Loftus, G. R., & Masson, M. E. J. (1994). Using confidence intervals in
within-subject designs. Psychonomic Bulletin & Review, 1, 476 – 490.
Logan, G. D., & Gordon, R. D. (2001). Executive control of visual
attention in dual-task situations. Psychological Review, 108, 393– 434.
Lukatela, G., & Turvey, M. T. (1994). Visual lexical access is initially
phonological: 2. Evidence from phonological priming by homophones
and pseudohomophones. Journal of Experimental Psychology: General,
123, 331–353.
MacLeod, C. M., & Masson, M. E. J. (2000). Repetition priming in
speeded word reading: Contributions of perceptual and conceptual processing episodes. Journal of Memory and Language, 42, 208 –228.
Mayr, U. (2001). Age differences in the selection of mental sets: The role
of inhibition, stimulus ambiguity, and response-set overlap. Psychology
and Aging, 16, 96 –109.
Mayr, U., & Keele, S. W. (2000). Changing internal constraints on action:
418
MASSON, BUB, WOODWARD, AND CHAN
The role of backward inhibition. Journal of Experimental Psychology:
General, 129, 4 –26.
Meiran, N. (1996). Reconfiguration of processing mode prior to task
performance. Journal of Experimental Psychology: Learning, Memory,
and Cognition, 22, 1423–1442.
Meiran, N. (2000). Modeling cognitive control in task-switching. Psychological Research, 63, 234 –249.
Meiran, N., Chorev, Z., & Sapir, A. (2000). Component processes in task
switching. Cognitive Psychology, 41, 211–253.
Monsell, S., Taylor, T. J., & Murphy, K. (2001). Naming the color of a
word: Is it responses or task sets that compete? Memory & Cognition, 29, 137–151.
Norman, D. A., & Shallice, T. (1986). Attention to action: Willed and
automatic control of behavior. In R. J. Davidson, G. E. Schwartz, & D.
Shapiro (Eds.), Consciousness and self-regulation (Vol. 4, pp. 1–18).
New York: Plenum.
Perfetti, C. A., & Bell, L. C. (1991). Phonemic activation during the first 40
ms of word identification: Evidence from backward masking and priming. Journal of Memory and Language, 30, 473– 485.
Rayner, K., Sereno, S. C., Lesch, M. F., & Pollatsek, A. (1995). Phonological codes are automatically activated during reading: Evidence from
an eye movement paradigm. Psychological Science, 6, 26 –32.
Rogers, R. D., & Monsell, S. (1995). Costs of a predictable switch between
simple cognitive tasks. Journal of Experimental Psychology: General,
124, 207–231.
Scarborough, D. L., Cortese, C., & Scarborough, H. S. (1977). Frequency
and repetition effects in lexical memory. Journal of Experimental Psychology: Human Perception and Performance, 3, 1–17.
Spector, A., & Biederman, I. (1976). Mental set and shift revisited. American Journal of Psychology, 89, 669 – 679.
Szymanski, K. F., & MacLeod, C. M. (1996). Manipulation of attention at
study affects an explicit but not an implicit test of memory. Consciousness and Cognition, 5, 165–175.
Taylor, D. A. (1976). Holistic and analytic processes in the comparison of
letters. Perception & Psychophysics, 20, 187–190.
Taylor, T. E., & Lupker, S. J. (2001). Sequential effects in naming: A
time-criterion account. Journal of Experimental Psychology: Learning,
Memory, and Cognition, 27, 117–138.
Ulrich, R., & Miller, J. (1994). Effects of truncation on reaction time
analysis. Journal of Experimental Psychology: General, 123, 34 – 80.
Waters, G. S., & Seidenberg, M. S. (1985). Spelling–sound effects in
reading: Time-course and decision criteria. Memory & Cognition, 13,
557–572.
Yee, P. L. (1991). Semantic inhibition of ignored words during a figure
classification task. Quarterly Journal of Experimental Psychology: Human Experimental Psychology, 43A, 127–153.
Received June 25, 2002
Revision received April 15, 2003
Accepted April 15, 2003 䡲